On this tenth day of the 12 Days of AI, we are exploring the bias that comes within AI systems, either intentionally or unintentionally. All systems have inherent bias because of the data sets that work beneath the interface. Some AI sites do a better job at countering bias than others.
The post reminds us:
“AI generative output is not creativity, but a statistical variation without intent and meaning. AI is a statistical artist. It introduces variations into its work, but its “choices” are driven on patterns and probabilities rather than deep understanding of lived experiences and emotions. AI generative outputs, while devoid of the intent and meaning characteristic of human creativity, possess significant potential to speed, support, and extend the creative process.”
The prompt was to go into an AI image generator, and use a prompt of a person in a setting, doing some action (they frame it as a scene from a movie) and look to see if there were any apparent bias in the results. I used Adobe Firefly and asked it to generate a “stern-looking school principal talking to an elementary teacher about student work.”
Is there bias in the two collections of results I received?
Out of the eight images, it seems as if five (and maybe six? Depends how you look at who is who) show a white male in the role of principal, and of the eight, only two images show non-white adults being generated. Two women seem to be in the administrative role. I didn’t have enough credits in Firefly to keep running the experiment a few more times, which would have been ideal. Would I have kept finding similar genders and races of the administrators?
What about if I refined my prompt, adding loaded terms like “urban school” or “struggling teacher” or “disruptive student”? How would other platforms perform with the same prompt? Would there be some gender/racial stereotypes into who is an administrator and who is an educator?
Peace (in explorations),
Kevin